Plant species recognition using leaf images and convolutional neural networks

Abstract

The identification of plant species is essential in botany and has attracted the interest of researchers in the field of computer science. Such identification requires the assistance of botany experts and, due to the substantial number of species and the similarity among them, it can be time-consuming and subjective. To automate the process of plant identification, computer systems that capture and process plant images have been considered. These systems use machine learning and therefore require image samples for training and model construction. Among the techniques that can be used for machine learning, convolutional neural networks have shown promise due to their ability to use images without prior preprocessing and background information. This work investigates the use of machine learning through convolutional neural networks to identifying plant species. For this, a new dataset of images from 35 plant species were created, collecting images from an arboreal collection, and, using data augmentation, this dataset was expanded. This dataset was used to evaluate the accuracies of four convolutional neural network models. The better accuracy value was equal to 89%, when using the MobileNetV2 model.